Application of neural networks for fault diagnosis in plant production

Sixth-generation computers, in Japan, have been announced as natural intelligence computers that would display behaviors based on biological rather than silicon models. These systems are derived from neurological models or neural networks that consist of a number of simple, highly interconnected processing elements which process information by their dynamic-state response to external inputs. The neural network is made by specifying interconnections, transfer functions, and training laws of the network. Then appropriate inputs are applied to the network, and it is allowed to react. The overall state of the network after it has reacted to the input will be the desired response pattern. This paper explores the potential for using simple networks for data interpretation. The ultimate goal of our study includes the interpretation of industrial plant monitoring data for diagnosis of chemical process or other process problems. Known data about normal operations, as well as abnormal operations and its causes, can be taught to the system. The system will apply a learning mechanism to the data in order to acquire knowledge. By supplying operational data, the system will indicate the nature of the fault, if any exists. 5 refs., 10 figs.